Optimization of Parameters of Cnn Based Method by Particle Swarm Optimization

Abstract

CNN based models are being developed for the analysis of medical images. These models are varying according to the structure of the tissue to be analyzed and the image acquisition technique. In our previous study, we developed a CNN-based model to perform automatic counting of follicles in the ovary. In the developed model, there are 3 basic parameters that affect segmentation success. These are General Stride (GS), Neighbor Distance (ND) and Patch Accuracy (PA), respectively. It is almost impossible to find the optimum values of these parameters manually. For this reason, in this study, parameter optimization of CNN based model was performed with Particle Swarm Optimization (PSO).As a result of the experimental studies, it was observed that the optimization of these 3 parameters increased the segmentation success of the model by 4.27%.